Safety manufacturing control method and device

Abstract

The invention discloses a safety manufacturing control method and device, relates to the technical field of intelligent manufacturing digitization, and constructs a set of intelligent monitoring and decision support system oriented to safety management of a manufacturing process through deep fusion of machine learning and a knowledge graph technology. And performing dynamic feature extraction and pattern recognition on the real-time data stream by using a machine learning model, and obtaining security disposal measures through a knowledge graph. The machine learning module is responsible for mining potential abnormal modes from data to predict safety risks in advance, the knowledge graph provides semantic association and causal logic support, multi-level and high-precision prediction of the safety risks is achieved through cooperation of the machine learning module and the knowledge graph, and the production safety of the manufacturing process is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing digitalization, in particular to a safety manufacturing control method and device based on machine learning and knowledge graph. BACKGROUND

[0002] Patent document CN118607930A describes a chemical safety risk management and control method and system based on a knowledge graph. The present application discloses a chemical safety risk management and control method and system based on a knowledge graph, which specifically includes the following modules: a chemical safety production data acquisition module, a chemical safety production legal regulations, standards and specifications, historical accident data acquisition module, a data preprocessing module, a knowledge graph module, a chemical safety risk situation analysis module, a dynamic updating and feedback module, and a chemical safety risk situation visualization module; S101: data acquisition; S102: data preprocessing: remove noise data and invalid data from the collected data; standardize different sources and types of data to have a unified measurement standard; store the processed data in a relational database or a NoSQL database; the chemical safety risk management and control method and system based on the knowledge graph can significantly enhance the chemical safety risk management and control.

[0003] Patent document CN120428564A describes a multi-model fusion collaborative energy-saving optimization control method and system based on a knowledge graph. The present application provides a multi-model fusion collaborative energy-saving optimization control method and system based on a knowledge graph, which is applied to the field of industrial equipment optimization control. The present application obtains target knowledge graph information and real-time data of target air conditioning system cooling demand, and the target knowledge graph information is used to represent entities, attributes and corresponding relationships matched with the target air conditioning system; the real-time data information of the target air conditioning system cooling demand is processed for feature extraction, generating real-time environmental parameter features and equipment operation state features; the real-time environmental parameter features and equipment operation state features are processed based on a target load prediction model, generating cooling demand information and load prediction information of a target time period; the target knowledge graph information, cooling demand information and load prediction information of the target time period are processed based on a target DQN model, generating global optimization control strategy information, which is used to adjust the system parameters of the target air conditioning system.

[0004] CN118607930A describes a chemical safety risk management and control method and system based on a knowledge graph. The present application is based on a dynamic chemical safety production risk knowledge graph, and performs risk situation analysis on real-time collected chemical safety production data. This method is difficult to implement. The present application is directed to the field of industrial safety production, and uses process key data as a detection target to identify safety risks through an algorithm model, and then queries a control strategy through a knowledge graph technology.

[0005] Other related conventional manufacturing process safety management technologies mainly rely on manual experience, lack real-time data monitoring and intelligent analysis capabilities, resulting in low efficiency and potential safety hazards. With the development of industrial digitization and intelligentization, how to use advanced technical means to realize real-time monitoring, prediction and control of potential risks in the manufacturing process has become a problem to be solved.

[0006] As can be seen, in the prior art, although there are some safety management methods based on data analysis, the following problems exist universally:

[0007] 1. Limited data processing capability, unable to efficiently analyze large-scale real-time data.

[0008] 2. Lack of intelligent risk prediction model, difficult to accurately identify potential safety hazards.

[0009] 3. Low system integration, difficult to seamlessly integrate with other industrial systems (such as data acquisition software, video monitoring system, etc.).

[0010] Therefore, there is an urgent need for a safety manufacturing control method and system based on advanced algorithms and knowledge graph to improve the safety and intelligent level of the manufacturing process. SUMMARY

[0011] In view of the above problems, the present application provides a safety manufacturing control method and device for overcoming the above problems or at least partially solving the above problems.

[0012] The present application provides the following solutions:

[0013] A safety manufacturing control method, comprising:

[0014] Constructing a safety risk prediction and evaluation model, the construction process comprising analyzing the historical data of the production line, extracting features from the preprocessed data, combining machine learning related algorithms to construct a key parameter prediction model; combining expert knowledge and process safety threshold to construct a risk assessment for the safety of the process corresponding to the production line, so as to perform safety risk assessment on key process parameters based on real-time data, real-time identify potential safety hazards, and output risk level;

[0015] Constructing a safety risk knowledge graph, the construction process comprising using knowledge graph technology to model the correlation between the equipment state and process parameter information of different processes in the manufacturing process, forming a structured knowledge base, so as to mine potential key knowledge and give safety risk analysis results and safety control strategy results;

[0016] A safety linkage control model is constructed, which adopts an adaptive fuzzy linkage control method, processes the nonlinearity and uncertainty of the process through semantic rules, supports continuous improvement based on expert knowledge, and realizes adaptive linkage feedback control.

[0017] Preferably, the method for constructing the key parameter prediction model comprises:

[0018] The key parameter data of the mixing process are collected, cleaned of missing values and outliers, standardized, and then converted into a sliding time window to form the model input format.

[0019] The CNN layer is designed to extract local features, the pooling layer is added to reduce dimensions, the LSTM layer is designed to capture time sequence features, the Dropout layer is added to prevent overfitting, the fully connected layer is constructed to output results, and the appropriate loss function and optimizer are selected. After multiple rounds of training, the model converges.

[0020] The RMSE index is calculated using the test set, and the preset threshold is compared to determine whether it meets the standard. If it does not meet the standard, the parameters of each layer are adjusted, and the process is repeated until the performance meets the standard and is deployed for actual multi-process key parameter prediction.

[0021] Preferably, the process safety risk assessment adopts a fuzzy multi-attribute decision-making method, which converts language variables into triangular fuzzy numbers with the help of fuzzy mathematics theory, combines multi-field expert knowledge and objective data characteristics, and realizes multi-dimensional trade-off in uncertain environment.

[0022] Preferably, the fuzzy decision framework is determined, and the key parameters of the mixing process temperature, mixing process speed, granulation process temperature, molding process pressure, and molding process temperature are recorded as the scheme set.

[0023] For each scheme set, evaluation indexes are designed based on key parameter prediction data, real-time data, safety thresholds, and other attributes, such as threshold deviation and variance.

[0024] The language term set describing the process risk is set to low, medium, and high to perform fuzzy evaluation on the performance of each scheme under each attribute.

[0025] Preferably, these language variables are converted into standardized triangular fuzzy numbers l, m, and u through the membership function, where l, m, and u represent the minimum possible value, the most possible value, and the maximum possible value of the evaluation, respectively.

[0026] Preferably, the individual evaluations are synthesized using a fuzzy aggregation operator to form a unified group fuzzy evaluation value x ij , x ij represents the comprehensive fuzzy score of scheme i under attribute j, and all group fuzzy evaluation values x ij are arranged in rows and columns to generate an m x n fuzzy decision matrix D.

[0027] Preferably: the attributes weights are determined by analytic hierarchy process or entropy weight method, and the weight sum is 1; the fuzzy comprehensive evaluation value of each process key parameter is calculated, the fuzzy evaluation under each attribute is weighted and fused to generate a fuzzy value reflecting the overall performance of the scheme, and the evaluation value is a triangular fuzzy number; the barycentric method is used to realize fuzzification to obtain the risk assessment of each process key parameter.

[0028] Preferably: the risk assessment indicators, and the material, process key parameters and control instruction time sequence are selected to construct a feature set;

[0029] The information gain of each feature is calculated, and the inherent value is introduced to standardize the information gain;

[0030] The binary split gain rate of each candidate point is calculated, and the optimal split point is selected to generate a binary branch;

[0031] If the data contains missing values, the samples are proportionally distributed to the child nodes according to the known values; when the samples of the current node are all of the same kind, there is no remaining feature available, or the number of samples is lower than the threshold, the recursion is terminated, at this time a leaf node is generated and the majority class is marked;

[0032] After the tree growth is completed, C4.5 performs post-pruning to control overfitting; pessimistic error pruning is used, the non-leaf nodes are traversed from bottom to top, the upper limit of node error rate is estimated by binomial distribution, and the estimated error rates before and after node pruning are compared; if the upper limit of the overall error rate after pruning is not higher than the upper limit of the sub-tree error rate, the leaf node is used to replace the sub-tree;

[0033] The C4.5 tree path is converted into IF-THEN rules, and redundant conditions are deleted;

[0034] The knowledge triplets are established, the semantic continuous feature split point, the quantized gain rate attribute and the retained pruning traces are used to convert the decision process into a traceable and verifiable multi-process safety risk knowledge network.

[0035] Preferably: the input and output variables are defined, the deviation of each process key parameter from the preset value is defined as an input variable, real-time monitoring data is defined as an output variable, the input variable threshold range is determined according to the safety threshold; the membership function topology structure is divided into 5 semantic subsets for each variable, and a triangle is used for non-linear mapping;

[0036] A fuzzy rule base is constructed, and IF-THEN rules are generated based on expert experience or data-driven methods;

[0037] When running online, the system fuzzifies the precise input value through the membership function, maps it into the membership degree of each subset, performs fuzzy reasoning, uses the minimum maximum operator for the antecedent, uses the TSK output linear function for the consequent, finally calculates the weighted center value of the membership degree distribution by using the barycenter method, and realizes the regulation and control of the key parameters of each process by de-fuzzifying the fuzzy output into the precise control quantity.

[0038] A safe manufacturing control device for executing the safe manufacturing control method described above, the device comprising:

[0039] A safe risk prediction and evaluation model construction unit for constructing a safe risk prediction and evaluation model, the construction process comprising feature extraction of preprocessed data by analyzing the production line historical data, combining machine learning related algorithms to construct a key parameter prediction model, combining expert knowledge and process safety threshold to construct a risk assessment of the corresponding process of the production line, so as to perform a safety risk assessment of key process parameters based on real-time data, real-time identification of potential safety hazards, and output of a risk level;

[0040] A safe risk knowledge graph construction unit for constructing a safe risk knowledge graph, the construction process comprising correlation modeling of the device state and process parameter information between different processes in the manufacturing process by using knowledge graph technology, forming a structured knowledge base, so as to mine potential key knowledge and give a safe risk analysis result and a safe control strategy result;

[0041] A safe linkage control unit for constructing a safe linkage control model, the safe linkage control model using an adaptive fuzzy linkage control method to process the nonlinearity and uncertainty of the process through semantic rules, supporting continuous improvement based on expert knowledge, and realizing adaptive linkage feedback control.

[0042] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0043] The safe manufacturing control method and device provided by the embodiments of the present application construct an intelligent monitoring and decision support system for manufacturing process safety management by deeply integrating machine learning and knowledge graph technology. The machine learning model is used to dynamically extract features and identify patterns from real-time data streams, and the knowledge graph is used to obtain safety disposal measures. The machine learning module is responsible for predicting safety risks in advance by mining potential abnormal patterns from data, and the knowledge graph provides semantic association and causal logic support. The two work together to realize multi-level and high-precision prediction of safety risks, effectively improving the safety of the manufacturing process.

[0044] By using a manufacturing safety knowledge graph, information such as equipment status and process parameters between different processes in the manufacturing process is correlated and modeled to form a structured knowledge base. Potential key knowledge is extracted, and risk analysis and control strategies are provided, which effectively improves the intelligence level of the system.

[0045] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0047] Figure 1 This is a flowchart of a safe manufacturing control method provided in an embodiment of the present invention;

[0048] Figure 2 This is an overall framework diagram of a safe manufacturing control method provided in an embodiment of the present invention;

[0049] Figure 3 This is a flowchart of the preprocessing of key process parameter data provided in the embodiments of the present invention;

[0050] Figure 4 This is a flowchart of the CNN-LSTM multi-stage key parameter prediction model processing provided in this embodiment of the invention;

[0051] Figure 5 This is a flowchart of risk assessment for key process parameters provided in an embodiment of the present invention;

[0052] Figure 6 This is a flowchart of the multi-process safety risk knowledge graph construction process provided in this embodiment of the invention;

[0053] Figure 7 This is a flowchart of the adaptive linkage control method provided in an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of a safe manufacturing control device provided in an embodiment of the present invention;

[0055] Figure 9 This is a schematic diagram of a safe manufacturing control device provided in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0057] See Figure 1 This invention provides a safe manufacturing control method, such as... Figure 1 As shown, the method may include:

[0058] S101: Constructing a safety risk prediction and assessment model. The construction process includes analyzing historical production line data, extracting features from preprocessed data, and combining machine learning algorithms to construct a key parameter prediction model; combining expert knowledge and process safety thresholds to construct a risk assessment of the corresponding process safety of the production line, so as to conduct safety risk assessment of key process parameters based on real-time data, identify potential safety hazards in real time, and output the risk level; in specific implementation, the method for constructing the key parameter prediction model in this application embodiment may include:

[0059] Collect key parameter data for mixed processes, clean up missing and outlier values, standardize the data, and then construct a sliding time window to convert it into the model input format.

[0060] The process involves sequentially designing CNN layers to extract local features, adding pooling layers for dimensionality reduction, designing LSTM layers to capture temporal features, adding Dropout layers to prevent overfitting, constructing fully connected layers to output results, selecting appropriate loss functions and optimizers, and training multiple times until convergence.

[0061] The RMSE metric is calculated using the test set and compared with the preset threshold to determine whether it meets the standard. If it does not meet the standard, the parameters of each layer are adjusted, and the process is repeated until the performance meets the standard before it is deployed for prediction of key parameters in actual multi-process operations.

[0062] The process safety risk assessment adopts a fuzzy multi-attribute decision-making method, which uses fuzzy mathematics theory to transform linguistic variables into triangular fuzzy numbers, and combines multi-domain expert knowledge with objective data characteristics to achieve multi-dimensional trade-offs in uncertain environments.

[0063] Furthermore, a fuzzy decision-making framework is established, and the key parameters of mixing process temperature, mixing process speed, granulation process temperature, molding process pressure, and molding process temperature are recorded as a set of schemes.

[0064] For each set of solutions, evaluation indicators are designed based on key parameter prediction data, real-time data, safety thresholds, etc., and attribute sets such as threshold deviation and variance are determined.

[0065] The set of language terms used by experts to characterize process risks is defined as low, medium, and high, in order to conduct a fuzzy evaluation of the performance of each option under each attribute.

[0066] These linguistic variables are transformed into standardized triangular fuzzy numbers l, m, u using membership functions, where l, m, u represent the minimum possible value, the most likely value, and the maximum possible value of the evaluation, respectively.

[0067] A fuzzy aggregation operator is used to synthesize individual evaluations to form a unified group fuzzy evaluation value x. ij x ij This represents the comprehensive fuzzy score of scheme i under attribute j, and represents the fuzzy evaluation values ​​x of all groups. ij Arrange the rows and columns to generate an m×n fuzzy decision matrix D.

[0068] Aggregation is performed by combining attribute weights, which are determined by analytic hierarchy process or entropy weight method, where the sum of weights is 1; the fuzzy comprehensive evaluation value of each key parameter of the process is calculated, and the fuzzy evaluations under each attribute are weighted and fused to generate a fuzzy value that reflects the overall performance of the scheme. This evaluation value is a triangular fuzzy number; the centroid method is used to achieve fuzzification and obtain the risk assessment of the key parameters of each process.

[0069] S102: Construct a safety risk knowledge graph. The construction process includes using knowledge graph technology to model the association between equipment status and process parameter information of different processes in the manufacturing process, forming a structured knowledge base to mine potential key knowledge and provide safety risk analysis results and safety control strategy results. In specific implementation, the embodiments of this application can provide the selection of risk assessment indicators, as well as material, process key parameters, and control command time series to construct a feature set.

[0070] Calculate the information gain for each feature, and standardize the information gain by introducing intrinsic values;

[0071] Calculate the binary splitting gain ratio for each candidate point, and select the optimal splitting point to generate a binary branch;

[0072] If the data contains missing values, the samples are allocated to the child nodes according to the proportion of known values. The recursion terminates when all samples in the current node belong to the same class, no remaining features are available, or the number of samples is lower than the threshold. At this time, leaf nodes are generated and the majority class is labeled.

[0073] After the tree growth is complete, C4.5 performs post-pruning to control overfitting; pessimistic error pruning is adopted, traversing non-leaf nodes from bottom to top, estimating the upper limit of the node error rate using a binomial distribution, and comparing the estimated error rate before and after node pruning; if the upper limit of the overall error rate after pruning is not higher than the upper limit of the subtree error rate, then the subtree is replaced with a leaf node;

[0074] Convert the C4.5 tree path to an IF-THEN rule and remove redundant conditions;

[0075] Establish knowledge triples by semantically representing continuous feature split points, quantifying gain rate attributes, and preserving pruning traces, so as to transform the decision-making process into a traceable and verifiable multi-process safety risk knowledge network.

[0076] S103: Construct a safety linkage control model. The safety linkage control model adopts an adaptive fuzzy linkage control method, which processes the nonlinearity and uncertainty of the process through semantic rules, supports continuous improvement based on expert knowledge, and realizes adaptive linkage feedback control.

[0077] In specific implementation, the embodiments of this application can provide the following: defining input and output variables, defining the deviation of key parameters of each process from preset values ​​as input variables, defining real-time monitoring data as output variables, determining the threshold range of input variables based on safety thresholds; membership function topology structure, dividing each variable into 5 semantic subsets, and using triangles for nonlinear mapping;

[0078] Construct a fuzzy rule base and generate IF-THEN rules based on expert experience or data-driven methods;

[0079] During online operation, the system fuzzifies the precise input values ​​through membership functions, mapping them to the membership degrees of each subset; performs fuzzy inference, using the min-max operator for the antecedent and the TSK output linear function for the consequent; finally, it uses the centroid method to calculate the weighted center value of the membership degree distribution, and defuzzifies the fuzzy output into precise control quantities, thereby enabling the regulation of key parameters in each process.

[0080] The safety manufacturing control method provided in this application analyzes historical data to construct a safety risk prediction and assessment model and a safety risk knowledge graph, and combines real-time production line data to carry out safety linkage control, thereby improving the safety of the industrial manufacturing process.

[0081] The safe manufacturing control method provided in the embodiments of this application will be described in detail below.

[0082] like Figure 2 As shown, this method employs key parameter prediction and safety risk assessment to identify safety risks and their levels in real time; utilizes knowledge graph technology to construct safety risk knowledge for multiple processes, thereby providing risk control strategy recommendations; and constructs a knowledge base and rule base to enable real-time reasoning and coordinated control.

[0083] 1. Data acquisition and preprocessing, such as Figure 3 As shown.

[0084] Data from different processes in the production line, such as the rotation speed, temperature, and viscosity in the kneading process, the temperature in the granulation process, and the pressure and temperature in the molding process, are key parameters. Real-time data may contain issues such as inaccuracies, outliers, and noise, which directly affect the performance, generalization ability, training efficiency, and reliability of the prediction model.

[0085] Key process parameter data preprocessing includes analyzing data characteristics such as mean, median, standard deviation, minimum, maximum, and quantiles. To address missing data, filtering and interpolation methods are used to ensure the integrity of the data sequence. For outliers, data points deviating from the normal range are identified, analyzed, corrected, or removed. These operations reduce the impact of noise, errors, or irrelevant information on the prediction model, improve its accuracy, and reduce the risk of overfitting.

[0086] 2. Construction of a safety risk prediction and assessment model.

[0087] By analyzing historical production line data, feature extraction is performed on the preprocessed data, and a key parameter prediction model is constructed by combining relevant machine learning algorithms. Risk assessment of production line process safety is conducted by combining expert knowledge and process safety thresholds, and safety risk assessment of key process parameters is performed based on real-time data to identify potential safety hazards in real time and output the risk level.

[0088] Construction of a key parameter prediction model.

[0089] Its core idea lies in combining the local feature extraction capability of CNN with the long-term temporal modeling capability of LSTM. First, the CNN convolutional neural network is used to extract the correlation of multiple process physical parameters, such as rotation speed and temperature, for key parameters of different processes in the production line, and to capture key local patterns. Then, the LSTM long short-term memory network is used to process the temporal dependence of key process parameters, solving the problems of long operation cycles, high dimensionality, and difficulty in temporal feature mining in multi-process operations.

[0090] CNN, acting as a front-end processor, automatically identifies short-term patterns in the input sequence through one-dimensional convolution operations, such as peak periods or fluctuation cycles in the time series of key parameters. LSTM, on the other hand, acts as a back-end time series analyzer, capturing the long-term dependencies between features extracted by CNN. This division of labor enables the algorithm to capture both the local spatial features of the data and model dynamic correlations across time, significantly improving the predictive ability for complex sequences with multiple processes.

[0091] The construction of a CNN-LSTM multi-process key parameter prediction model is as follows: Figure 4 As shown, the specific steps are as follows:

[0092] (1) Data collection and preprocessing. Collect key parameter data of mixed processes, clean missing and outlier values, standardize, and then construct a sliding time window to convert it into the model input format.

[0093] (2) Construct and train the CNN-LSTM model. Design CNN layers to extract local features, add pooling layers to reduce dimensionality, design LSTM layers to capture temporal features, add Dropout layers to prevent overfitting, construct fully connected layers to output results, select appropriate loss functions and optimizers, and train multiple times until convergence.

[0094] (3) Evaluation and optimization of the model. Calculate indicators such as RMSE using the test set, compare them with the preset thresholds to determine whether they meet the standards. If they do not meet the standards, adjust the parameters of each layer, repeat until the performance meets the standards, and then deploy the model for prediction of key parameters in actual multi-process operations.

[0095] Process safety risk assessment, such as Figure 5 As shown.

[0096] Risk warning for mixing, granulation, and molding processes is a crucial guarantee for safe production, and process safety risk assessment is an important technical support. Real-world decision-making often faces information ambiguity, diverse risk attributes, and dynamic environments. Employing a fuzzy multi-attribute decision-making method, which uses fuzzy mathematics theory to transform linguistic variables into triangular fuzzy numbers and combines multi-domain expert knowledge with objective data characteristics, can achieve multi-dimensional trade-offs in uncertain environments, improving the scientific rigor and adaptability of complex decisions regarding multiple process key parameters.

[0097] The process fuzzy evaluation matrix is ​​constructed as follows: First, a fuzzy decision-making framework is determined, and key parameters such as mixing process temperature, mixing process speed, granulation process temperature, molding process pressure, and molding process temperature are denoted as a set of schemes. For each scheme set, evaluation indicators are designed based on predicted data of key parameters, real-time data, and safety thresholds, and attribute sets such as threshold deviation and variance are determined. The set of linguistic terms used by experts to characterize process risks is defined as ({low, medium, high}) to perform fuzzy evaluation of the performance of each scheme under each attribute. Furthermore, these linguistic variables are transformed into standardized fuzzy numbers (triangular fuzzy numbers (l,m,u)) through membership functions. Here, l, m, and u represent the minimum possible value, the most likely value, and the maximum possible value of the evaluation, respectively.

[0098] To avoid decision-making bias from individual experts, it is necessary to integrate group evaluations. A fuzzy aggregation operator is used to synthesize individual evaluations, forming a unified group fuzzy evaluation value x. ij (x ij This represents the comprehensive fuzzy score of scheme i under attribute j. Finally, all x... ij Arrange the rows and columns to generate an m×n fuzzy decision matrix D, which lays the foundation for subsequent weighted decision-making.

[0099] Fuzzy risk assessment of the process. First, aggregation is performed by combining attribute weights, which are determined by the analytic hierarchy process (AHP) or entropy weight method, where the sum of the weights is 1. The fuzzy comprehensive evaluation value of each key parameter of the process is calculated, and the fuzzy evaluations under each attribute are weighted and fused to generate a fuzzy value reflecting the overall performance of the solution; this evaluation value is the triangular fuzzy number. Then, the centroid method is used to achieve fuzzification, obtaining the risk assessment of the key parameters of each process step.

[0100] 3. Construction of a multi-process safety risk knowledge graph, such as... Figure 6 As shown.

[0101] By leveraging knowledge graph technology, the system models the relationships between equipment status, process parameters, and other information across different manufacturing processes, forming a structured knowledge base. This allows for the discovery of potential key knowledge and the provision of risk analysis and control strategies. For example, the system can record potential safety hazards that may arise during a particular process under specific temperature and pressure conditions and provide corresponding control strategies.

[0102] To mine knowledge information between key parameter time-series data and obtain multi-process safety risk knowledge representations, this project uses C4.5 to mine the association rules between various elements. First, risk assessments (low, medium, high) and time series data of materials, process key parameters, and control instructions are selected to construct a feature set. The information gain of each feature is calculated, and intrinsic values ​​are introduced to standardize the information gain, where intrinsic values ​​reflect the dispersion of feature values. The binary splitting gain ratio of each candidate point is calculated, and the optimal split point is selected to generate a binary branch. If the data contains missing values, samples are allocated to child nodes according to the proportion of known values. Recursion terminates when all samples in the current node belong to the same class, no remaining features are available, or the number of samples is below a threshold. At this point, leaf nodes are generated and labeled as the majority class.

[0103] After tree growth is complete, C4.5 performs post-pruning to control overfitting. Pessimistic error pruning is employed: non-leaf nodes are traversed from bottom to top, and the upper limit of the node error rate is estimated using a binomial distribution. The estimated error rate before and after pruning is compared. If the overall upper limit of the error rate after pruning is not higher than the upper limit of the subtree error rate, the subtree is replaced with a leaf node (the leaf node's class is determined by the majority class of the subtree samples). This process balances model complexity and accuracy, ultimately generating a more concise and robust decision tree. Finally, the C4.5 tree path is converted to IF-THEN rules, and redundant conditions are removed to further improve interpretability.

[0104] Furthermore, a knowledge triplet (key parameters, risk assessment, control strategy) is established. By semanticizing continuous feature split points, quantifying gain rate attributes, and preserving pruning traces, the above decision-making process is transformed into a traceable and verifiable multi-process safety risk knowledge network.

[0105] 4. Safety linkage control, such asFigure 7 As shown.

[0106] Stable operation of key parameters in processes such as mixing, granulation, and molding is crucial for product quality. However, these key parameters are characterized by unknown precise models, nonlinearity, and uncertainty, making them difficult to control using model-driven methods. Therefore, an adaptive fuzzy linkage control method is adopted. By handling the nonlinearity and uncertainty of the processes through semantic rules and leveraging the open architecture of its rule base, it supports continuous improvement based on expert knowledge, achieving adaptive linkage feedback control and enhancing the reliability and stability of each process.

[0107] Knowledge modeling and rule base construction.

[0108] First, input and output variables are defined. The deviations of key parameters from preset values ​​in each process are defined as input variables, and real-time monitoring data is defined as output variables. Based on safety thresholds, the threshold ranges for input variables are determined. Further, the membership function topology is defined, dividing each variable into five semantic subsets (NB negative large, NS negative small, Z zero, PS positive small, PB positive large), using a triangular nonlinear mapping. Then, a fuzzy rule base is constructed, typically generating "IF-THEN" rules (e.g., "IF deviation is negative large, THEN output is medium") based on expert experience or data-driven methods. These rules cover all typical operating conditions, and the membership function parameters can be further adjusted and optimized to ensure the rule base is both complete and adaptable.

[0109] Real-time reasoning and control decision-making.

[0110] During online operation, the system fuzzifies the precise input values ​​using membership functions, mapping them to the membership degrees of each subset. Fuzzy inference is then performed, with the antecedent using the min-max operator and the consequent using a TSK output linear function. Finally, the centroid method is used to calculate the weighted center value of the membership degree distribution. Defuzzification transforms the fuzzy output into precise control quantities, enabling the regulation of key parameters in each process (data acquisition, transmission, and control parameters).

[0111] As can be seen, the safety manufacturing control method provided in this application, through the deep integration of machine learning and knowledge graph technologies, constructs an intelligent monitoring and decision support system for manufacturing process safety management. Compared with the closest existing technology, the method provided in this application brings significant improvements in several aspects, specifically reflected in the following:

[0112] 1. Significantly improves data processing capabilities and risk prediction accuracy:

[0113] Existing technologies largely rely on structured rules or single data sources, making it difficult to handle multi-source, heterogeneous, and high-dimensional real-time data in manufacturing environments (such as equipment sensor data, operation logs, and environmental parameters). The method provided in this application utilizes machine learning models to dynamically extract features and recognize patterns in real-time data streams, while simultaneously acquiring safety mitigation measures through a knowledge graph. The machine learning module is responsible for mining potential anomaly patterns from the data to predict safety risks in advance, while the knowledge graph provides semantic association and causal logic support. Together, they achieve multi-level, high-precision prediction of safety risks, effectively improving production safety in the manufacturing process.

[0114] 2. Achieve systematic management and intelligent reasoning of manufacturing safety knowledge through knowledge graphs:

[0115] Traditional systems lack the ability to structurally integrate and dynamically evolve safety knowledge, often leading to knowledge isolation and delayed response. The manufacturing safety knowledge graph constructed in this application models the correlation between equipment status, process parameters, and other information across different processes in manufacturing, forming a structured knowledge base. This allows for the discovery of potential key knowledge and the provision of risk analysis and control strategies, effectively improving the system's intelligence level.

[0116] In summary, the method provided in this application, through the synergistic innovation of machine learning and knowledge graphs, has achieved a paradigm shift in manufacturing safety management from "passive response" to "proactive prediction" and from "experience-driven" to "knowledge-driven," providing a reliable technical path for the intelligent upgrading of industrial safety management.

[0117] See Figure 8 This application embodiment can also provide a safe manufacturing control device, such as... Figure 8 As shown, the apparatus for performing the above-described safe manufacturing control method may include:

[0118] The safety risk prediction and assessment model construction unit 801 is used to construct a safety risk prediction and assessment model. The construction process includes analyzing historical data of the production line, extracting features from the preprocessed data, and combining machine learning algorithms to construct a key parameter prediction model; combining expert knowledge and process safety thresholds to construct a risk assessment of the safety of the corresponding process of the production line, so as to conduct safety risk assessment of key process parameters based on real-time data, identify potential safety hazards in real time, and output the risk level.

[0119] The safety risk knowledge graph construction unit 802 is used to construct a safety risk knowledge graph. The construction process includes using knowledge graph technology to model the association between equipment status and process parameter information of different processes in the manufacturing process, forming a structured knowledge base, so as to mine potential key knowledge and provide safety risk analysis results and safety control strategy results.

[0120] The safety linkage control unit 803 is used to construct a safety linkage control model. The safety linkage control model is used to adopt an adaptive fuzzy linkage control method, which processes the nonlinearity and uncertainty of the process through semantic rules, supports continuous improvement based on expert knowledge, and realizes adaptive linkage feedback control.

[0121] This application embodiment can also provide a safe manufacturing control device, the device including a processor and a memory:

[0122] The memory is used to store program code and transmit the program code to the processor;

[0123] The processor is used to execute the steps of the above-described safe manufacturing control method according to the instructions in the program code.

[0124] like Figure 9 As shown in the illustration, a safe manufacturing control device provided in this application embodiment may include: a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, memory 11, and communication interface 12 all communicate with each other through the communication bus 13.

[0125] In this embodiment, the processor 10 may be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0126] The processor 10 can call programs stored in the memory 11. Specifically, the processor 10 can execute operations in the embodiments of the safe manufacturing control method.

[0127] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions:

[0128] The construction of a safety risk prediction and assessment model includes analyzing historical production line data, extracting features from preprocessed data, and combining machine learning algorithms to build a key parameter prediction model. It also involves combining expert knowledge and process safety thresholds to conduct a risk assessment of the corresponding process safety of the production line, so as to conduct a safety risk assessment of key process parameters based on real-time data, identify potential safety hazards in real time, and output the risk level.

[0129] Constructing a safety risk knowledge graph involves using knowledge graph technology to model the associations between equipment status and process parameters of different processes in the manufacturing process, forming a structured knowledge base to uncover potential key knowledge and provide safety risk analysis results and safety control strategy results.

[0130] A safety linkage control model is constructed. This model is used to employ an adaptive fuzzy linkage control method, which processes the nonlinearity and uncertainty of the process through semantic rules, supports continuous improvement based on expert knowledge, and realizes adaptive linkage feedback control.

[0131] In one possible implementation, the memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function (such as file creation or data read / write). The data storage area may store data created during use, such as initialization data.

[0132] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0133] Communication interface 12 can be an interface for a communication model, used to connect with other devices or systems.

[0134] Of course, it should be noted that, Figure 9 The structure shown does not constitute a limitation on the safety manufacturing control device in the embodiments of this application. In practical applications, the safety manufacturing control device may include more than Figure 9 More or fewer components as shown, or combinations of certain components.

[0135] This application embodiment may also provide a computer-readable storage medium for storing program code for executing the steps of the above-described safe manufacturing control method.

[0136] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0137] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0138] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0139] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A safety manufacturing control method characterized by, include: The construction of a safety risk prediction and assessment model includes analyzing historical production line data, extracting features from preprocessed data, and combining machine learning algorithms to build a key parameter prediction model. It also involves combining expert knowledge and process safety thresholds to conduct a risk assessment of the corresponding process safety of the production line, so as to conduct a safety risk assessment of key process parameters based on real-time data, identify potential safety hazards in real time, and output the risk level. Constructing a safety risk knowledge graph involves using knowledge graph technology to model the associations between equipment status and process parameters of different processes in the manufacturing process, forming a structured knowledge base to uncover potential key knowledge and provide safety risk analysis results and safety control strategy results. A safety linkage control model is constructed, which adopts an adaptive fuzzy linkage control method. The nonlinearity and uncertainty of the process are handled through semantic rules, and the model supports continuous improvement based on expert knowledge to achieve adaptive linkage feedback control.

2. The safety manufacturing control method according to claim 1, characterized by, The method for constructing the key parameter prediction model includes: Collect key parameter data for mixed processes, clean up missing and outlier values, standardize the data, and then construct a sliding time window to convert it into the model input format. The process involves sequentially designing CNN layers to extract local features, adding pooling layers for dimensionality reduction, designing LSTM layers to capture temporal features, adding Dropout layers to prevent overfitting, constructing fully connected layers to output results, selecting appropriate loss functions and optimizers, and training multiple times until convergence. The RMSE metric is calculated using the test set and compared with the preset threshold to determine whether it meets the standard. If it does not meet the standard, the parameters of each layer are adjusted, and the process is repeated until the performance meets the standard before it is deployed for prediction of key parameters in actual multi-process operations.

3. The safety manufacturing control method according to claim 1, characterized by, The process safety risk assessment adopts a fuzzy multi-attribute decision-making method, which uses fuzzy mathematics theory to transform linguistic variables into triangular fuzzy numbers, and combines multi-domain expert knowledge with objective data characteristics to achieve multi-dimensional trade-offs in uncertain environments.

4. The safety manufacturing control method according to claim 3, characterized by, A fuzzy decision-making framework is established, and the key parameters of mixing process temperature, mixing process speed, granulation process temperature, molding process pressure, and molding process temperature are recorded as a set of schemes. For each set of solutions, evaluation indicators are designed based on key parameter prediction data, real-time data, safety thresholds, etc., and attribute sets such as threshold deviation and variance are determined. The set of language terms used by experts to characterize process risks is defined as low, medium, and high, in order to conduct a fuzzy evaluation of the performance of each option under each attribute.

5. The safety manufacturing control method according to claim 4, characterized by, These linguistic variables are transformed into standardized triangular fuzzy numbers l, m, u using membership functions, where l, m, u represent the minimum possible value, the most likely value, and the maximum possible value of the evaluation, respectively.

6. The safety manufacturing control method according to claim 3, characterized by, The individual evaluations are synthesized by using fuzzy aggregation operators to form a unified group fuzzy evaluation value x ij , x ij represents the comprehensive fuzzy evaluation score of scheme i at attribute j, and all group fuzzy evaluation values x ij are arranged in rows and columns to generate an m x n fuzzy decision matrix D.

7. The safety manufacturing control method according to claim 3, characterized by, Aggregation is performed by combining attribute weights, which are determined by analytic hierarchy process or entropy weight method, where the sum of weights is 1; the fuzzy comprehensive evaluation value of each key parameter of the process is calculated, and the fuzzy evaluations under each attribute are weighted and fused to generate a fuzzy value that reflects the overall performance of the scheme. This evaluation value is a triangular fuzzy number; the centroid method is used to achieve fuzzification and obtain the risk assessment of the key parameters of each process.

8. The safety manufacturing control method according to claim 1, characterized by, Select risk assessment indicators, as well as key parameters of materials and processes, and time series of control instructions to construct a feature set; The information gain of each feature is calculated, and the eigenvalue is introduced to normalize the information gain; The binary split gain rate of each candidate point is calculated, and the optimal split point is selected to generate a binary branch; If the data contains missing values, the samples are distributed to the child nodes according to the proportion of known values; when the samples of the current node are all of the same kind, there is no remaining feature available, or the number of samples is lower than the threshold, the recursion is terminated, and a leaf node is generated and marked as the majority class; After the tree growth is completed, C4.5 performs post-pruning to control overfitting; the pessimistic error pruning is adopted, the non-leaf nodes are traversed from bottom to top, the upper limit of the node error rate is estimated by using the binomial distribution, and the estimated error rates before and after pruning are compared; if the upper limit of the overall error rate after pruning is not higher than the upper limit of the sub-tree error rate, the leaf node is used to replace the sub-tree; The C4.5 tree path is converted into an IF-THEN rule, and redundant conditions are deleted; The knowledge triplets are established, the semantic continuous feature split point, the quantified gain rate attribute and the pruning traces are retained, so as to convert the decision process into a traceable and verifiable multi-process safety risk knowledge network.

9. The safety manufacturing control method according to claim 1, characterized by, The input and output variables are defined, the deviation of each process key parameter from the preset value is defined as an input variable, real-time monitoring data is defined as an output variable, and the input variable threshold range is determined according to the safety threshold; The membership function topology structure is divided into five semantic subsets for each variable, and a triangular shape is used for nonlinear mapping; A fuzzy rule base is constructed, and IF-THEN rules are generated based on expert experience or data-driven methods; During online operation, the system fuzzifies the accurate input value through the membership function, maps it to the membership degree of each subset, executes fuzzy reasoning, uses the minimum-maximum operator for the antecedent, uses the TSK output linear function for the consequent, finally calculates the weighted center value of the membership degree distribution by using the barycenter method, and de-fuzzifies the fuzzy output to convert it into an accurate control quantity, so as to realize the regulation and control of each process key parameter.

10. A safety manufacturing control apparatus characterized by comprising: The device is used to execute the safety manufacturing control method in any one of claims 1-9, and the device comprises: a safety risk prediction and evaluation model construction unit, configured to construct a safety risk prediction and evaluation model, and the construction process comprises: performing feature extraction on preprocessed data by analyzing production line historical data, combining machine learning related algorithms, and constructing a key parameter prediction model; combining expert knowledge and process safety threshold to construct a risk evaluation model for process safety of the production line, so as to perform safety risk evaluation on key process parameters based on real-time data, identify potential safety hazards in real time, and output a risk level; a safety risk knowledge graph construction unit, configured to construct a safety risk knowledge graph, and the construction process comprises: using knowledge graph technology to correlate and model device states and process parameter information between different processes in the manufacturing process, and forming a structured knowledge base, so as to mine potential key knowledge and give safety risk analysis results and safety control strategy results; a safety linkage control unit, configured to construct a safety linkage control model, and the safety linkage control model adopts an adaptive fuzzy linkage control method, processes the nonlinearity and uncertainty of the process through semantic rules, supports continuous improvement based on expert knowledge, and realizes adaptive linkage feedback control.

Images